Collaborative Research: High Performance Multi-Scale Ocean Modelling

该项目将为高性能、高度并行的计算机开发和实施新的算法，以增强罗格斯大学开发的光谱元素海洋模型。作为该算法开发的测试平台，生成的代码将应用于多尺度全球海洋模拟，该模拟将侧重于东北太平洋大陆架，包括美国西海岸和阿拉斯加沿海湾。这些地区是一个调查项目的中心，旨在了解多种时空尺度的物理过程如何影响海洋生态系统。所开发的新算法将解决与长期和多尺度全球海洋模拟相关的计算困难。研究将集中在鲁棒的二维和三维椭圆迭代求解器，基于缓存的处理器上的代码优化，以及非结构化网格上的可扩展并行性能。特别是，它将开发一种新的处理快速屋檐的方法，避免分裂正压模式，从而从模型中消除浅水方程。

This project will develop and implement new algorithms for high performance, highly parallel computers to enhance the Spectral Element Ocean Model developed at Rutgers University. As a testbed for this algorithmic development, the code produced will be applied to multi-scale global oceanic simulations which will focus on the Northeast Pacific continental shelves, including the Western Coast of the United States and the Coastal Gulf of Alaska. These regions are at the center of an investigative program aimed at understanding how physical processes, with multiple temporal and spatial scales, influence marine ecosystems. The new algorithms developed will address the computational difficulties associates with the long and multi-scale global oceanic simulations. The research will focus on robust two-and three-dimensional elliptic iterative solvers, optimization of the code on cache-based processors, and scalable parallel performance on unstructured grids. In particular it will exploit a new treatment for fast eaves that avoids splitting the barotropic mode and, thus, eliminates the shallow water equations from the model.